Eur Biophys J (2000) 29: 439±454
Ó Springer-Verlag 2000
ARTICLE
Mafalda Nina á Simon BerneÁche á BenoõÃ t Roux
Anchoring of a monotopic membrane protein: the binding of prostaglandin H2 synthase-1 to the surface of a phospholipid bilayer
Received: 20 December 1999 / Revised version: 26 March 2000 / Accepted: 26 March 2000
Abstract Prostaglandin H2 synthases (PGHS-1 and -2) are monotopic peripheral membrane proteins that catalyse the synthesis of prostaglandins in the arachidonate cascade. Picot et al. (1994) proposed that the enzyme is anchored to one lea¯et of the bilayer by a membrane anchoring domain consisting of a right-handed spiral of amphipathic helices (residues 73±116) forming a planar motif. Two dierent computational approaches are used to examine the association of the PGHS-1 membrane anchoring domain with a membrane via the proposed mechanism. The electrostatic contribution to the free energy of solvation is obtained by solving numerically the ®nite-dierence Poisson equation for the protein attached to a membrane represented as a planar slab of low dielectric. The nonpolar cavity formation and van der Waals contributions to the solvation free energy are assumed to be proportional to the water accessible surface area. Based on the optimum position determined from the continuum solvent model, two atomic models of the PGHS-1 anchoring domain associated with an explicit dimyristoylphosphatidylcholine (DMPC) bilayer diering by the thickness of the membrane bilayer were constructed. A total of 2 ns molecular dynamics simulation were performed to study the details of lipidprotein interactions at the microscopic level. In the simulations the lipid hydrocarbon chains interacting with the anchoring domain assume various shapes, suggesting that the plasticity of the membrane is significant. The hydrophobic residues in the membrane side of the helices interact with the hydrophobic membrane M. Nina1 á S. BerneÁche2 á B. Roux (&)2 Membrane Transport Research Group (GRTM), Department of Chemistry, University of Montreal, C.P. 6128, succ. A, Quebec H3C 3J7, Canada e-mail:
[email protected] Current addresses: 1 Laboratoire de Biologie Structurale, I.G.B.M.C., 4 Rue Laurent Fries, 67404 Illkirch Cedex, Strasbourg, France 2 Department of Biochemistry and Structural Biology, Weill Medical College of Cornell University, New York, NY 10021, USA
core, while the positively charged residues interact with the lipid polar headgroups to stabilize the anchoring of the membrane domain to the upper half of the bilayer. The phosphate headgroup of one DMPC molecule disposed at the center of the spiral formed by helices A, B, C and D interacts strongly with Arg120, a residue on helix D that has previously been identi®ed as being important in the activity of PGHS-1. In the full enzyme structure, this position corresponds to the entrance of a long hydrophobic channel leading to the cyclooxygenase active site. These observations provide insights into the association of the arachidonic acid substrate to the cyclooxygenase active site of PGHS-1. Key words Prostaglandin H2 synthase-1 á Phospholipid membranes á Computer simulation á Poisson equation á Mean-®eld approximation
Introduction Prostaglandin H2 synthases (PGHS) are N-glycosylated membrane proteins responsible for the synthesis of prostaglandins involved in important physiological processes such as smooth muscle contraction, in¯ammation, parturition and platelet aggregation (Otto et al. 1993; Smith et al. 1991). Two isoforms have been identi®ed: PGHS-1 and PGHS-2, also referred as cyclooxygenase COX-1 and COX-2 (for reviews, see Hla et al. 1993; Loll and Garavito 1994). PGHS-1 is constitutively expressed in most tissues and is responsible for the physiological production of prostaglandins. PGHS-2 is induced by cytokines, mitogens and endotoxins in in¯ammatory cells, and is responsible for the elevated production of prostaglandins during in¯ammation. Prostaglandin H2 synthases are located primarily in the endoplasmic reticulum (Smith and Marnett 1991) and possess two catalytic active sites to perform the synthesis of prostaglandin H2 from arachidonic acid in two consecutive steps: a cyclooxygenase site that converts
440
arachidonic acid to PGG2 and a hydroperoxydase site that reduces the hydroxyl group of PGG2 to PGH2 (Ohki et al. 1979; van der Ouderaa et al. 1977; Pagels et al. 1983). The ®rst product, PGG2 hydroperoxide, is converted into one of several prostanoid hormones. The cyclooxygenase activity of PGHS-1 and PGHS-2 is the target of an economically important class of drugs known as non-steroidal anti-in¯ammatory drugs (NSAIDs), which makes it one of the most interesting targets in the pharmaceutical industry (Holtzman et al. 1991; Meade et al. 1993). The primary mechanism of NSAIDs in the treatment of in¯ammation is the inhibition of the two isoforms of PGHS. Much recent eort has been made to produce selective inhibitors of PGHS2 in the belief that these will lack the gastrointestinal damaging eects of traditional NSAIDs (McCartney et al. 1999). The three-dimensional structures of ovine 576-amino acid PGHS-1 complexed with the NSAID ¯urbiprofen (Picot et al. 1994), with 2-bromoacetoxybenzoic acid (a potent aspirin analogue) (Loll et al. 1995), as well as with iodinated indomethacin and suprofen (Loll et al. 1996), have been determined by X-ray crystallography. X-ray structures of unliganded murine PGHS-2 and complexed with ¯urbiprofen, indomethacin and a selective inhibitor (SC-558) have also been obtained (Kurumbail et al. 1996). The structure of human PGHS-2, determined in the presence of a selective inhibitor, is similar to that of PGHS-1 (Luong et al. 1996). PGHS are classi®ed as integral membrane proteins because detergents are required to extract the enzyme from the membrane (Smith and Marnett 1991). However, in contrast to many other membrane proteins such as the photosynthetic reaction center (Deisenhofer and Michel 1989), bacteriorhodopsin (Henderson et al. 1990), porins (Cowan et al. 1995) or the KcsA K+ channel from Streptomyces lividans (Doyle et al. 1998), the crystal structure of PGHS-1 reveals no transmembrane motifs which could serve to anchor the protein to the lipid membrane (see Fig. 1) (Picot et al. 1994). Since it is unlikely that the native conformation is signi®cantly aected by detergent extraction or crystallization, a novel and attractive model for membrane attachment was proposed by Picot et al. (1994) based on a close examination of the X-ray structure. According to this model, the dimer is anchored to one lea¯et of the bilayer through the hydrophobic surface of a number of amphipathic helices, in a similar way that surface-active peptides bind to membranes with their nonpolar residues on the lower surfaces interacting with the hydrophobic interior of the bilayer (Kaiser and Kezdy 1987). The proposed membrane attachment motif consists of a right-handed spiral of amphipathic a-helices A, B, C and D (residues 73±116) lying approximately in a plane acting as a membrane anchoring domain (see Fig. 1). A corresponding motif on the adjacent monomer faces in the same direction. Helices A, B, C and the beginning of helix D are amphipathic and their hydrophobic surface faces outward away from the main body of the
Fig. 1 PGHS-1 dimer viewed along the dimer interface with the molecular two-fold axis vertical (Picot et al. 1994). The catalytic domain, the EGF-like module and the anchoring domain are colored in blue, green and orange, respectively
protein, forming a large hydrophobic patch on the exterior of the dimer (Fig. 2). The amphipathic helices A (residues 73±82), B (residues 86±92), C (residues 97±105) and the beginning of helix D (residues 108±116) contain hydrophobic residues such as alanines, valines, leucines, isoleucines and, interestingly, several aromatic residues, such as phenylalanines (4) and tryptophans (3). The opposite side of the helices contains polar and a few positively charged residues (mainly arginines) which could interact with the lipid polar headgroups. Based on the crystal structures of the PGHS-1 and PGHS-2, it was con®rmed by experiments on chimeric proteins which did not contain the putative membrane anchoring domains that the amphipathic helices act as membrane anchors (Yi et al. 1998). Recent results by Spencer et al. (1999) provide direct support for the hypothesis that PGHS-1 is a monotopic membrane protein which associates with bilayers through the anchoring domain. Membrane-associated forms of ovine PGHS-1 and human PGHS-2 were labeled using a hydrophobic, photoactivable reagent, isolated, cleaved and the photolabeled peptides were sequenced. The results provide direct biochemical support for the hypothesis that PGHS-1 and -2 do associate with membranes through the monotopic anchoring domain. Furthermore, it was demonstrated, via site directed mutations, that the amphipathic character of each helix (A, B, and C) is important for the assembly and folding of ovine PGHS-1 to a cyclooxygenase active form. Mutants, in which three or four hydrophobic residues of the helices expected to protrude into the membrane, were replaced with small neutral residues. These mutants (often misfolded aggregates) exhibited little or no catalytic activity when expressed in COS-1 cells. Interestingly, a similar mode of membrane anchoring via amphipathic helices has been recently suggested in the case of squalene cyclase (Wendt et al. 1997, 1999). PGHS and squalene cyclase share structurally similar
441
Fig. 2 DINO/MSMS (Philippsen 1999; Sanner et al. 1995) representation of the molecular surface of the PGHS-1 dimer. The hydrophobic residues surface is colored in yellow and the hydrophilic residues surface is red. Two viewing directions parallel and perpendicular to the molecular two-fold axis
membrane anchoring motifs but have little or no sequence similarity. The membrane anchoring region of the squalene cyclase forms a ¯at nonpolar plateau (consisting of one a-helix, one loop and one segment between two a-helices) and encircled by a ring of positive charged residues (Wendt et al. 1997, 1999). According to the proposed model, PGHS-1 is a monotopic membrane protein since the binding motif does not extend beyond one lea¯et of the bilayer. Thus, PGHS-1 and squalene cyclase are interesting from a structural point of view since they provide the ®rst high-resolution structures of peripheral membrane proteins (Loll et al. 1995; Picot et al. 1994; Wendt et al. 1997). However, although the X-ray structure is suggestive, the detailed mode of interaction involved in the stable anchoring of PGHS cannot be elucidated based on the available experimental data. There is not enough information at the present time to completely determine the location and orientation of the membrane-
bound form of the protein relative to the bilayer surface. A better understanding of the factors responsible for the interaction of PGHS-1 with a membrane would provide some insight into the attachment mechanism of peripheral membrane proteins in general. In addition, the proposed model has also some important functional implications for the cyclooxygenase activity of PGHS1. The four-helices binding motif forms the entrance to a long narrow hydrophobic channel extending from the membrane surface and leading to the cyclooxygenase active site (Picot et al. 1994), which could be the pathway for the entry of arachidonic acid, the substrate of the cyclooxygenase activity. Unlike PGHS, such a polar pathway is not required for the function of squalene cyclase because its substrate (squalene) is entirely nonpolar (Wendt et al. 1997, 1999). Arachidonic acid is principally produced by phospholipidase A2 (Smith et al. 1991). It is released into the hydrocarbon core of the lipid bilayer, where it is likely to remain because of its hydrophobicity. However, its movement from the membrane to the active site lying at the top of the hydrophobic channel has not been elucidated. A possible mechanism is that arachidonic acid penetrates directly into the cyclooxygenase active site from the interior of the bilayer through helices A, B and C of the anchoring domain without going into the bulk solvent region. The positively charged arginines of the anchoring domain may contribute to facilitate the movement of arachidonic acid in the channel leading to the cyclooxygenase active site through an interaction of the Arg120 guanidinium group with the arachidonate carboxylate group. It has been suggested that the carbonyl group of arachidonic acid interacts with Arg120 and Tyr385 and may determine the stereoselectivity of PGHS-1 for inhibitors (Bhattacharyya et al. 1996). Mutagenesis studies con®rmed that the Arg120 residue of PGHS-1 is critical for binding the substrate and inhibitors through ionic interactions of its guanidinium group with the carboxylate moieties of arachidonic acid and some NSAIDs (Bhattacharyya et al. 1996; Mancini et al. 1995; Rieke et al. 1999). In view of its structural as well as functional implications, it is important to better characterize the membrane attachment of PGHS-1. The goal of this paper is to examine the microscopic basis of the model proposed by Picot et al. (1994) for the anchoring of PGHS-1 to a membrane using computational methods. Two complementary approaches were used. First, the optimal con®guration of the anchoring domain relative to the membrane was examined with implicit continuum models used to describe the protein solvation free energy. Second, two detailed atomic models of the anchoring domain bound to a solvated dimyristoylphosphatidylcholine (DMPC) bilayer were constructed and simulated with molecular dynamics (MD). The theoretical methods employed are described in the next section. Then the results are described and discussed. The paper is concluded with a brief summary of the main results.
442
Theory and methods Mean-®eld continuum solvation model Solvation eects provide the major force driving the association of proteins to membranes (Ben-Tal et al. 1996; BerneÁche et al. 1998; Edholm and Jahnig 1988; Eisenberg and McLachlan 1986; Milik and Skolnick 1993). To investigate the importance of various energetic factors in the membrane association of the anchoring domain of PGHS-1, a mean-®eld potential based on a continuum approximation was used. A similar approach has been used previous by BenTal et al. (1996) and BerneÁche et al. (1998). The total free energy of solvation DGtot is decomposed into a nonpolar cavity formation DGnp and an electrostatic contribution DGelec (Boresch et al. 1994; Gilson and Honig 1988): DGtot DGnp DGelec
1
The term DGnp accounts for the nonpolar contributions and is assumed to be related to the water accessible surface area of the protein. Assuming that the membrane normal is oriented along the Z-axis, a simple sum over the water accessible surface area of all atoms is used to approximate the nonpolar contribution: X np DGnp
Z Si cf
zi
2 i
where zi is the position of the ith atom along the Z-axis, Snp i is the water accessible surface area of the ith atom, and c 33 cal mol)1 AÊ)2 is the surface tension coecient obtained from experimental free energies of the transfer of hydrocarbons from the pure liquid alkane to water (Hermann 1972). The function f(z) models the transition from pure water to pure alkane at the interface:
jzj Z0 2 =DZ 2 if jzj Z0 f
z e
3 1 otherwise where Z0 and DZ are the width of the core region and the width of the core region-polar heads interface, taken to be 10.0 and 2.5 AÊ, respectively, according to experimental data on lipid bilayers (Jacobs and White 1989; White and Wiener 1996). The water Ê accessible surface area Snp i was calculated by rolling a 1.4 A probe on the protein van der Waals surface while atomic radii were taken from the all-hydrogen PARM22 force ®eld (Mackerell et al. 1998). The term DGelec accounts for the reaction ®eld contribution to the free energy of the peptide in the nonuniform dielectric media. The electrostatic contribution to the free energy of transfer from water to the membrane along the Z-axis was computed using the Poisson equation for macroscopic continuum electrostatics (Honig and Nicholls 1995; Honig et al. 1993; Warwicker and Watson 1982): r
rr/
r
4pqprot
r
4 prot
where (r) is the position-dependent dielectric constant and q (r) is the charge density due to the protein. The protein was represented at the microscopic level with its associated atomic radii and atomic charges. All individual helices and the anchoring domain were treated as neutral. Speci®cally, blocking groups preserving the adjacent atoms of the amide plane were used to preserve the neutrality of the helical segments (CH3-CO- for N-terminus and NH-CH3 for C-terminus). The atomic charges were taken from the all-hydrogen PARM22 force ®eld (Mackerell et al. 1998); the atomic radii used to de®ne the protein-solvent dielectric interface were derived from radial solvent charge distribution functions of the explicit solvent around the 20 standard amino acids (Nina et al. 1997). It was shown previously that this approach is able to yield a quantitative representation of the electrostatic contribution to the solvation free energies of amino acids that is consistent with thermodynamics perturbation techniques and MD simulations with an explicit model of the solvent (Nina et al. 1997). In the continuum electrostatic calculations, the membrane was represented by a planar slab of 25 AÊ thick corresponding to the width of the hydrocarbon core of the membrane (White and Wiener 1996). Dielectric constants were assigned according to the polarity of the
medium: 80 for bulk water, 2 for the membrane and 1 for the protein. Because of the uncertainty on the continuum description, no intermediate dielectric region was assigned to the water-lipid interface and the region corresponding to the polar headgroups was assumed to have a dielectric constant of 80. All calculations were performed with a cubic grid of 80 AÊ with two grid points per AÊ. The membrane geometrical center was translated to )15 AÊ along the Z-axis of a three-dimensional cubic grid. The ionic strength of the surrounding bulk solution was set to zero (no counterions were included). The protein was mapped onto the grid with its backbone center of mass placed at a distance Z from the geometrical center of the membrane. For each value of Z, the electrostatic contribution to the free energy of transfer from the water to the membrane was calculated by subtracting the electrostatic energy computed in a continuous medium representative of water from the electrostatic energy computed in a membrane immersed in a solvent region: 1 X memb qi /
ri /bulk
ri
5 DGelec 2 i where qi is the charge of the ith atom in the protein, /memb(ri) is the total electrostatic potential of the protein near the membrane, /bulk(ri) is the total electrostatic potential of the protein immersed in the bulk region [/(r) is the solution of the Poisson equation]. Calculations were performed using the ®nite-dierence algorithm of Klapper et al. (1986), as implemented in the PBEQ facility (Beglov D, Im W, Roux B, unpublished) of the biomolecular simulation program CHARMM (Brooks et al. 1983). Construction of the microscopic models The membrane-bound three-dimensional conformation of the PGHS-1 anchoring domain was modeled using the 3.5 AÊ resolution X-ray structure available from the Protein Data Bank (Picot et al. 1994), and modi®ed as described in the section Results and discussion. Only a single monomer was considered in the current computations to limit the number of atoms. Blocking groups preserving the adjacent atoms of the amide plane were used for the Nterminus (CH3-CO-) and the C-terminus (-NH-CH3). The atomic model used for the MD simulations is composed of the 54 amino acids anchoring domain of one PGHS-1 monomer (helices A, B, C and D), 71 DMPC molecules and 2938 water molecules, for a total of 18,129 atoms. This constitutes the central unit of a periodic system, the dimensions of which are 48 ´ 56 AÊ2 in the xy directions. The membrane normal was oriented along the Z-axis. The two-fold crystallographic dimer axis (Picot et al. 1994) was assumed to be parallel to the bilayer normal and the corresponding orientation of the monomeric anchoring domain was preserved. There are 29 and 42 lipids in the upper and lower halves of the bilayer, respectively. The four helix motif was anchored in the upper part of the membrane. Periodic boundary conditions were applied in the xy directions to mimic an in®nite planar bilayer. A translation distance corresponding to the elementary box size along the Z-axis was used along the Z-axis to simulate a periodic multilayer system. The potential energy function used for the calculations was the all-hydrogen PARM22 force ®eld (Mackerell et al. 1998) of the biomolecular CHARMM program (Brooks et al. 1983) which includes lipid molecules (Schlenkrich et al. 1996). The TIP3P model was used to represent the water molecules (Jorgensen et al. 1983). The hydrogen positions of the protein were constructed using the CHARMM subroutine HBUILD (Brunger and Karplus 1988). Adopted Newton-Raphson basis and steepest descent algorithms were used for the energy minimizations (Brooks et al. 1983). A careful construction of such complex system is required in order to obtain a starting con®guration that is representative of the solvated protein-membrane system. The general protocol developed by Woolf and Roux (1994, 1996) was used to construct the initial model. Such method has been used previously to generate other protein-membrane systems such as the gramicidin channel (Woolf and Roux 1994, 1996), bacteriophage Pf1 (Roux and Woolf 1996) and melittin in lipid bilayers (BerneÁche et al. 1998). Since the
443 membrane-associated PGHS-1 anchoring domain is lying parallel to the surface of the bilayer, it is necessary to account for the crosssectional area of the protein to determine the appropriate number of DMPC molecules to include in the upper and lower parts of the bilayer. The total cross-sectional area of the anchoring domain with its center of mass disposed at Z 0 was determined as a function of Z. The cross-sectional area of the anchoring domain (not shown) varies from 100 AÊ2 to a maximum of 780 AÊ2 along the bilayer normal. The maximum in the cross-sectional area corresponds to the three amphipathic helices (A, B and C) lying parallel to the membrane plane and forming part of the membrane anchoring domain. The larger cross-sectional area of the protein corresponds roughly to that of 13 DMPC molecules. Assuming that the anchoring domain is bound to the upper halve of the bilayer, the simulation systems were constructed with 29 and 42 DMPC molecules in the upper and the lower halves of the bilayer, respectively, to accommodate the cross-sectional area of the PGHS-1 anchoring domain. To determine the initial position of each lipid relative to the protein, the eective lipid particles were modeled as a large Lennard-Jones sphere with a cross-sectional area of 64 AÊ2, corresponding to the average cross-sectional area of one DMPC molecule in the liquid-crystalline La phase (Gennis 1989; Nagle 1993). Their packing around the protein was determined from a MD simulation in which the particles in both halves were restrained along Z 17 AÊ. The resulting spheres con®gurations were then replaced by full pre-equilibrated and pre-hydrated DMPC molecules chosen randomly from a library of 2000 DMPC molecules (Loof et al. 1991; Pastor et al. 1991; Venable et al. 1993). In this library, each phospholipid polar headgroup is pre-hydrated with approximately 20 water molecules. The number of bad contacts and atomic overlaps were reduced by performing systematic rigid-body rotations of the lipids around the Z-axis and translations in the xyplane. One DMPC molecule of the upper half was manually disposed in the channel entrance formed by helices A, B, C and D. The total cross-sectional area of the entire DMPC/PGHS-1 system is 2690 AÊ2. Finally, the system was fully hydrated by overlaying a preequilibrated box of water molecules of the appropriate dimension to obtain about 35 water molecules per lipid. Two protein-membrane systems (I and II) were constructed with dierent membrane thickness. In system I, the thickness of the bilayer was kept identical to that of a pure lipid membrane. The dimension of the central unit of the periodic system is 48 AÊ ´ 56 AÊ ´ 68.4 AÊ and the center of the bilayer is at Z 0. In system II, the membrane thickness was reduced to account for the presence of the anchoring domain. Because of the presence of the anchoring domain, there is a large vacuum created in the upper half of the bilayer with the current construction protocol. Such a large defect is unrealistic. In a fully relaxed and equilibrated system the thickness of the upper layer may be expected to be signi®cantly perturbed by the presence of the protein. To deduce the thickness of the upper layer, we assume that the density from the carbon acyl chains is roughly similar in both the upper and lower halves of the bilayer. The density is n/V, where n is the number of DMPC molecules and V is the volume of the lipid core region of a layer. The volume of a layer is equal to the area of the bilayer times the thickness of the core region for that layer. The thickness of the lower layer, constituted of 42 lipids and no protein, was set to 12 AÊ in accord with scattering data (White and Wiener 1996). The density of the layer is 42/12 3.5 lipids AÊ)1. Thus, the thickness of the upper layer constituted of 29 lipids should be equal to 29/3.5 8.3 AÊ. To yield a similar density in both layers the bilayer of system II was assembled with the upper headgroups of the upper and lower halves located at Z )13.3 AÊ and +17 AÊ, respectively. The dimension of the central unit of the periodic system II is 48 AÊ ´ 56 AÊ ´ 64.7 AÊ and the center of the bilayer is at Z +1.85 AÊ. Equilibration and simulation procedures Two atomic systems of PGHS-1 anchored to a bilayer diering in the thickness of the membrane were equilibrated and simulated
(systems I and II). The trajectories were calculated in the microcanonical ensemble with a constant number of atoms, volume and energy (NVE). The average temperature of the system was set to 330 K, above the gel-liquid phase transition of DMPC (Gennis 1989). The list of nonbonded interactions was truncated at 12 AÊ using a group-based cuto. The electrostatic and van der Waals interactions were smoothly switched to zero from 8±11 AÊ. Since our main interest is to examine the structural reorganization in the hydrocarbon core of the membrane which rapidly takes place in response to the presence of the anchoring domain, we do not expect that the truncation of the nonbonded interactions has an impact on the current observations. The equations of motion were integrated with a time-step of 2 fs. The length of all bonds involving hydrogen atoms was ®xed with the SHAKE algorithm (Ryckaert et al. 1977). The center of mass of the protein was kept at the center of the membrane plane in the xy-plane. In the initial stage, the protein was ®xed in place to prevent large spurious motions. In addition, various energy restraints were used to ensure a smooth transition of the atomic systems toward a relaxed con®guration. Planar harmonic restraints were applied to the center of mass of the lipid polar headgroups (at Z 17 AÊ for system I and Z )13.3 AÊ and +17 AÊ for system II) to maintain the planarity of the membrane. Planar repulsive potentials were used to prevent excessive penetration of water molecules into the hydrocarbon region (within |Z|
